Table 4.

Summary of predictive models for NMOSD and multiple sclerosis based on MRI.

					Weighted Means
Model (Ref)	TP	FP	TN	FN	TP Rate	FP Rate	Precision	F-measure	ROC Area
Paty (29)	29	9	91	33	0.637	0.259	0.752	0.669	0.689
Swanton (30)	48	43	57	14	0.696	0.352	0.633	0.642	0.672
Matthews/Juryńczyk/Hyun (14, 31, 33)	34	25	75	28	0.626	0.324	0.634	0.630	0.649
Liao (18)	5	0	100	57	0.432	4.352	0.861	0.390	0.540
Bensi (32)	33	24	76	29	0.619	0.327	0.634	0.626	0.646
NMO/MS score	53	9	91	9	0.876	0.111	0.876	0.876	0.882
Machine learning algorithm	52	9	91	10	0.866	0.117	0.871	0.868	0.874

Criteria definitions.

Paty = 3 or more white matter lesions or if one periventricular lesion, 3 or more white matter lesions indicates MS, otherwise NMOSD.

Swanton = at least 2 of the following criteria: (1) periventricular lesion; (2) juxtacortical lesion; (3) infratentorial lesion; and (4) spinal cord lesion indicates MS, otherwise NMOSD.

Matthews = lesion adjacent to lateral body of ventricle (periventricular lesion) and inferior temporal lobe lesion, and U-fibre lesion (juxtracortical lesion) or Dawson's finger type lesion indicates MS, otherwise NMOSD.

Liao = linear ependymal lesion (pencil corpus callosum, hypothalamic, periaqueductal, periventricular, third ventricle) and not meeting Matthews criteria indicates NMSOD, otherwise MS.

NMOD/MS Score = NMOSD Score × 3.5 > MS score indicates NMSOD, otherwise MS.

NMOSD Score = counting 1 for each of Gd-enhancing, whole (axial) cord, swollen, central or atrophic lesion spinal cord lesions; Gd-enhancing, T2, chiasm or longitudinally extensive lesion optic nerve lesions; and nucleus tractus solitarius, periaqueductal, third ventricular, hypothalamic, leptomeningeal Gd-enhancing, central medullary, cloud-like or area postrema brain lesions, and counting 2 for longitudinally extensive spinal cord lesions and bilateral optic nerve lesions.

MS Score = counting 1 for each of short segment lesion and partial cord lesions, and ovoid, Dawson's fingers, pyramidal corpus callosum, other corpus callosum, periventricular, temporal lobe, splenium, cerebellar, large supratentorial, cortical, subcortical, juxtacortical, cerebellar peduncle T2 lesion, T1 black holes, brain Gd-enhancing lesion, and 9 or more T2 brain lesions.

Machine Learning Algorithm = see Figure 11 for algorithm.

Data is presented as ability to predict NMOSD vs. multiple sclerosis. Rates, precision and F-measure are all weighted average figures for predicting NMOSD and multiple sclerosis (i.e., overall accuracy).

Ref, reference; TP, true positive; FP, false positive; TN, true negative; FN, false negative; ROC, receiver operator characteristic.